Yong Wang

FASA: FREQUENCY-AWARE SPARSE ATTENTION

Tree Search for LLM Agent Reinforcement Learning

Everything in Its Place: Benchmarking Spatial Intelligence of Text-to-Image Models

BiasScope: Towards Automated Detection of Bias in LLM-as-a-Judge Evaluation

Decomposing Policy Optimization into Proxy Objective and Knowledge Distillation for Long-context Reasoning

Don't Lose Sight: Visually-Grounded Credit Assignment for Multimodal Reasoning

MMGenBench: Fully Automatically Evaluating LMMs from the Text-to-Image Generation Perspective

GPG: A Simple and Strong Reinforcement Learning Baseline for Model Reasoning

Harder Is Better: Boosting Mathematical Reasoning via Difficulty-Aware GRPO and Multi-Aspect Question Reformulation

Acceleration-Aware Sampling for Few-Step Rectified Flow Models

Direct Alignment of Language Models via Quality-Aware Self-Refinement